A picture I keep finding in inherited accounts: optimization score 95+, auto-apply on, and the account structure unrecognisable — audiences expanded, match types loosened, budget recommendations scattered across campaigns. When I ask, the answer is always the same: "Google recommends it, so we apply it. Our score is high." There's a fundamental frame to fix here.
Whose objective are recommendations optimized for?
Google Ads recommendations aren't malicious; but they aren't neutral either. What the system suggests usually shares a common denominator: more reach, more automation, more spending flexibility. These always align with Google's revenue model; they only sometimes align with your profitability. The "raise your budget, you're hitting the cap" suggestion never asks the marginal ROAS question from my previous article — it's true you're capped; whether the traffic beyond that cap earns you money is simply not the recommendation's concern.
My three-way classification
Usually apply: technical hygiene. Fixing disapproved ads, conflicting negative keywords, missing assets/extensions, conversion tracking issues, removed landing pages. No debate — this is the account's plumbing.
Evaluate case by case: smart bidding transitions, new campaign types (PMax included), audience expansions, "add similar keywords" suggestions. These are strategy decisions and are never applied without context. A smart-bidding suggestion is premature on an account with thin conversion data and spot-on for a strong one. The same recommendation produces opposite outcomes on two accounts.
Reject by default: broad-match migrations, budget increases, "expand your reach" prompts — and above all their automatic application. Broad match is a powerful tool when used deliberately with smart bidding and a solid negative list; migrated in bulk via a recommend-and-apply button, it floods your search-terms report with irrelevant queries.
Auto-apply: the most expensive checkbox
Riskier than the Recommendations tab itself is the setting that lets recommendations apply automatically. Leaving it on means handing structural decisions — match types, audiences, sometimes ad copy — to an unsupervised process. The bad practice I see at agencies: auto-apply left on at account creation, and months later nobody can answer "who made this change?" That's why the Change History screen is part of my weekly routine: I want to know the origin of every change I didn't make.
Fixing the relationship with optimization score
Optimization score is not a health score; it's a recommendation-acceptance score. There are two ways to raise it: apply a recommendation, or dismiss it with a reason — yes, dismissing raises the score too. That design alone tells you what the score measures: not your account's profitability, but the tidiness of the recommendation queue. My accounts usually sit in the 80–90 band, deliberately; a score of 100 most often means "every suggestion was accepted unquestioned." If you audit accounts on the client or hiring side, treat a 100-score account with suspicion, not admiration.
In short: read the Recommendations tab like a consultant, not a boss. You listen to a consultant, test against context, and take what works. Apply the plumbing, weigh the strategy, test the spending suggestions against your own arithmetic — and turn auto-apply off.